Repository navigation
skill-quality / semantic-skill-quality times out at its 10-minute limit (≈40% of runs cancelled, incl. main) #5228
Description
Parent Outcome: #5097
Work level: Delivery · Track: Active delivery · Intended Status: Ready
Owner
Root/fdev assigns.
Done when
semantic-skill-quality (skill-quality.yml, timeout-minutes: 10) stops being cancelled on its timeout.
- The problem: passing runs take 6–9 minutes. Today about 6 of the last 15 runs were cancelled, including main at 16:48Z, 18:21Z and 21:05Z, and PRs Registry: coord-engine and telemetry-host package-version lag the feeds (0.113.0 / 0.8.0 vs 0.114.3 / 0.9.0); feed check red on main #5216, ALGO-01.1 — ADR: ownership, taxonomy, knowledge-pack format and product-workspace distribution #5223 and atlas: record the accepted standard input pipeline decision, FS.GG.Game ADR 0001 (Refs FS-GG/FS.GG.Game#787) #5221. Each failed run costs a rerun on a 20-job-capped org.
- Measure: measure where the time goes.
- Fix: make it faster, for example with caching, by narrowing scope to changed skills, or by sharding. If it must stay slow, set an evidence-based timeout with headroom.
- Proof: 10 consecutive runs complete without a timeout, and the wall time is recorded.
- Required status: decide whether a timeout-cancelled
semantic-skill-qualitycounts as failure or neutral for merges, and document it.
This is a process finding reported by the GAME-RENDER-02.1 owner (#5221) and the ALGO-01.1 owner (#5227).
Current deliverable
Source preparation on routine/5228-semantic-gate-runtime; no PR yet.
Next action
Assigned gate_5228 owner measures semantic rejection-step cost and prepares a bounded optimization; Root admits a PR when integration capacity frees.
Blocked on
Nothing.
Acceptance evidence
Original cancelled run and measured phase timing: #5228 (comment) . Ten successful runs and required-status disposition remain unfulfilled.
Refs #5182
Activity
EHotwagner commentedon Nov 2, 2026
Root recovery finding, 2026-11-02: the inherited PR #5244 rerun also ended cancelled, run https://github.com/FS-GG/.github/actions/runs/41409628030 . Native job timestamps: 08:14:26–08:24:41 UTC; setup finished 08:14:47; compiled contracts/process checks finished 08:16:25; semantic quality/rejection-path step ran 08:16:25–08:24:39 and was cancelled. This adds a measured affected run to the existing finding, not a new issue or an accepted check. Root keeps #5244 draft, owns resolving this shared gate before claiming its handoff ready condition, and will reconsider implementation when the current .github PR integration capacity frees. No unchanged rerun dispatched.
EHotwagner commentedon Nov 2, 2026
EHotwagner commentedon Nov 2, 2026
Root coordination update, 2026-11-02: the existing reviewed two-file optimization at
e85d319df9784ada03f184c3bd78e06714f95177is being composed into the same Core upgrade PR #5255 by its existing owner algo_deps37. This avoids a competing source PR for the same actual bottleneck. The original optimization owner/review evidence remains retained.Actual original Core skill-quality run41457678933/job125318448487 exceeded its existing ten-minute limit. The retained job annotation identifies the timeout; no manual cancellation or superseding run was established. The board-mutant harness consumed323.711 seconds before later prerequisite controls continued to the deadline. Root authorized this causal repair: fail fast only after a mutant is detected, preserving the full unmutated baseline, surviving-mutant suites, all73 mutants and all60 rejection controls. No deadline or coverage relaxation is authorized. The assembled modified fixture and independent delta review are required before the next admitted hosted window.
This is source composition, not acceptance of this Delivery. Its full ten-consecutive-hosted-success/timing/status acceptance remains open. Native source integration, each original failed/cancelled attempt and the existing parent scope remain distinct.